Artificial Intelligence

We focus on solving AI’s most challenging issues, from building systems that automatically adapt to novel situations to creating AI that explains its own reasoning.

ARTIFICIAL INTELLIGENCE

We focus on solving AI’s most challenging issues, from building systems that automatically adapt to novel situations to creating AI that explains its own reasoning.

Human-Centric AI

Our advances in AI are improving medical training and care, helping save endangered species, and protecting our country. As leaders in the field of explainability, we are answering the global call to make AI systems trustworthy and accountable. At Charles River, we believe AI systems are intended to serve people: to amplify their capabilities, protect their interests and well-being, and support—not replace—their decision making. To achieve these goals and maximize the potential benefits of AI, we focus on creating AI-human collaborative systems.

A Leading Laboratory

Charles River’s scientists and engineers have been conducting leading-edge research since our founding nearly 40 years ago. Our open, collegial lab collaborates with dozens of universities and research labs across the U.S. We are currently engaged in more than 200 R&D projects, focusing on some of the biggest challenges in AI. Find out more from these selected academic publications and presentations.

Brittle AI, Causal Confusion, and Bad Mental Models: Challenges and Successes in the XAI Program

The authors discuss their key findings from the DARPA XAI program, which sought to “open the black box” of AI systems without sacrificing performance, and suggest future work in the field of explainable AI.

Towards Incorporating Artificial Intelligence in the Mission Planning Process

The authors describe the application of joint cognitive system analysis and neural policy programs to create effective mission plans as well as a method for prototyping these approaches based on the StarCraft II strategy game.

Probabilistic Programming:
Past, Present, and Future

In this invited keynote, Dr. Avi Pfeffer describes the unique capabilities of probabilistic programming languages (PPLs) and the development of two PPLs at Charles River: Figaro™ and Scruff™.

Or view the complete list of AI publications from Charles River researchers.

Solutions That Work

We combine our in-house research with our deep insights from working with teams in the real world to produce AI solutions that work the way you expect them to. Our projects and products open new frontiers for the Government and business sectors, improving operations through reduced costs, increased efficiency, and enhanced performance of both humans and machines.

Explainable AI

Scientists and engineers at Charles River Analytics are generating new knowledge at the frontiers of this rapidly growing area, creating a future where humans and AIs collaborate—to drive cars, respond to disasters, and make medical diagnoses.

Learn more about our XAI work

Machine Learning and Neural Networks

Our machine learning R&D identifies patterns in data using a variety of approaches, including reinforcement learning, deep learning, anomaly detection, and cluster analysis. We use these patterns to better understand the world and to make better predictions and decisions.

datavoid — DATAVOID
**Data analysis tool to assess value of information for decision making**

MERLIN uses a meta-reinforcement learning approach so swarms can autonomously learn new tactics — MERLIN-RST
**Increase swarm effectiveness with tactics derived from meta-reinforcement learning**

coltrain — COLTRANE
**A system that uses machine learning and symbolic reasoning to handle novelty in open-world domains**

Explainable AI from Charles River Analytics will enhance mission planning and operations for the Navy and Marine Corps. — EXTRA
**Tools that make DRL agents understandable and trustable**

solar — SOLAR
**Mixed reality application suite for space situational awareness**

As autonomous vehicles become more crucial on the battlefield, the CAMEL project is enhancing their value by developing explainable AI (XAI). (US Army photo by Patrick Albright) — CAMEL
**Advancing the field of explainable AI (XAI) with causal models to explain learning**

Probabilistic Programming

By enhancing programming languages with probabilistic capabilities, we are able to create systems that can help make decisions in the face of uncertainty. Scientists and engineers at Charles River have been leaders in the field of probabilistic programming since its earliest days. Our current work includes developing probabilistic programming languages (PPLs) and applying PPLs to specific challenges, such as missile defense, space object classification, and the health of engineered systems.

SNAPPR
**Optimize preventive maintenance with real-time estimates of operational ability**

ProbSysM — PROBSYSML
**Probabilistic extensions for the Systems Modeling Language**

Two US Army soldiers looking at computer monitors showing images of robotic combat vehicles. — RAPS
**A maintenance system that can predict the health and status of robotic combat vehicles**

SCRUFF-header_SM2 — Scruff
**A framework for building artificial intelligence systems**

powered — POWERED
**An application that uses probabilistic modeling to determine the reliability of power supplies**

MAAGI
**Identify malware attacks faster by recognizing code and techniques from previous attacks**

Dr. Pfeffer authored Practical Probabilistic Programming, which introduces readers to probabilistic programming and helps modelers without experience in machine learning create rich, probabilistic modeling applications. — PLEDGES
**A new machine learning approach with probabilistic modeling**

Optimization and Automated Planning

Optimizing actions and plans is a key feature of decision-support tools, especially when decision makers have multiple goals. At Charles River, we help our customers face the challenge of creating multi-step plans that adapt to the real world. We use a variety of algorithms, including genetic/evolutionary algorithms, market-based negotiation algorithms, neural networks, and ant-colony optimization, to develop tools for optimization and automated planning. By integrating these algorithms with our expertise in designing human-computer interfaces, we create solutions that not only generate effective and efficient plans but also help decision makers understand their options and the potential tradeoffs.

MIN-COST
**Use AI to adapt USV covertness mode based on current conditions**

CANDID
**Detect, mitigate, and prevent cybersecurity attacks on military ground vehicles.**

INTUIT
**Role and task-tailored interfaces for mission planning**

COMPASS
**A weather and climate forecasting tool that facilitates mission planning**

IMPRINT
**A virtual reality system for training first responders**

MANTA
**A platform-independent, natural control, and autonomy robot appliqué**

DATEM
**Monitor health and status of critical systems with machine-learning technologies**

Natural Language Processing

We use natural language understanding and natural language generation to strengthen the connection between humans and computers. We develop software that reads text documents and maps them to semantic representations for automated reasoning, and we incorporate natural language processing into a wide variety of applications, including intuitive interfaces for interacting with robots and uncrewed vehicles, tools for analyzing propaganda, and automated methods to predict cyberattacks.

Computer Vision and Perception

Our work on computer vision goes beyond the traditional definition of enabling computers to see, identify, and process images the way people do. We apply AI and machine learning to create BVLOS (beyond visual line of sight) capabilities and to fuse optical data with infrared, radiation, and other sensor data. We also build software tools that enable developers to use our advanced technologies to prototype their own computer vision systems.

ROMAN
**Real-time sea surface maps and maritime object detection**

VISION-kit — VISIONKIT
**A library of computer vision processing tools**

CROWSNEST
**A maritime traffic awareness solution for an uncrewed surface vessel**

ORION
**Human-computer interfaces for US Army commanders**

TNT-SCAN uses its network of cameras and radiation sensors to track a threat as it moves through a monitored area, providing data to security personnel that lets them take action quickly. — TNT-SCAN
**A networked system that tracks nuclear threats in urban environments**

follow-me — Follow-Me
**Enables ground robots to autonomously track and follow a designated leader**

Behavioral and Cognitive Modeling

Through our research, we seek to understand how the human brain works and then develop behavioral and cognitive models that can be used to predict or simulate human behavior. These models inform a variety of our capabilities, including performance assessment, intelligent tutoring, adversary behavior modeling, and cultural training.

Social Intelligence

Social intelligence enables AI systems to infer the goals and situational knowledge of their human partners, predict their needs, and offer context-aware actions. Our work in this area builds on many years of experience with sociocultural modeling, using key insights from the social sciences and reflecting the unique decision-support needs of the operational community.